@selfage/datastore_client

Install

npm install @selfage/datastore_client

Overview

Written in TypeScript and compiled to ES6 with inline source map & source. See @selfage/tsconfig for full compiler options. Provides type-safe Google Cloud Datastore APIs as a thin layer on top of @google-cloud/datastore, though requiring DatastoreModelDescriptors and QueryBuilders generated from @selfage/generator_cli.

You are encouraged to understand how Datastore works essentially before using this lib.

Example generated code

See @selfage/generator_cli#datastore-client for how to generate DatastoreModelDescriptors and QueryBuilders. Suppose the following has been generated and committed as task_model.ts. We will continue using the example below.

import { DatastoreQuery, DatastoreFilter, DatastoreModelDescriptor } from '@selfage/datastore_client/model_descriptor';
import { Task, TASK } from './task'; // Also generated from @selfage/generator_cli.

export let TASK_MODEL: DatastoreModelDescriptor<Task> = {
  name: "Task",
  key: "id",
  excludedIndexes: ["id", "payload"],
  valueDescriptor: TASK,
}

export class TaskDoneQueryBuilder {
  private datastoreQuery: DatastoreQuery<Task> = {
    modelDescriptor: TASK_MODEL,
    filters: new Array<DatastoreFilter>(),
    orderings: [
      {
        fieldName: "created",
        descending: true
      },
      {
        fieldName: "priority",
        descending: false
      },
    ]
  };

  public start(cursor: string): this {
    this.datastoreQuery.startCursor = cursor;
    return this;
  }
  public limit(num: number): this {
    this.datastoreQuery.limit = num;
    return this;
  }
  public equalToDone(value: boolean): this {
    this.datastoreQuery.filters.push({
      fieldName: "done",
      fieldValue: value,
      operator: "=",
    });
    return this;
  }
  public build(): DatastoreQuery<Task> {
    return this.datastoreQuery;
  }
}

export class TaskDoneSinceQueryBuilder {
  private datastoreQuery: DatastoreQuery<Task> = {
    modelDescriptor: TASK_MODEL,
    filters: new Array<DatastoreFilter>(),
    orderings: [
      {
        fieldName: "created",
        descending: true
      },
      {
        fieldName: "priority",
        descending: false
      },
    ]
  };

  public start(cursor: string): this {
    this.datastoreQuery.startCursor = cursor;
    return this;
  }
  public limit(num: number): this {
    this.datastoreQuery.limit = num;
    return this;
  }
  public equalToDone(value: boolean): this {
    this.datastoreQuery.filters.push({
      fieldName: "done",
      fieldValue: value,
      operator: "=",
    });
    return this;
  }
  public greaterThanCreated(value: number): this {
    this.datastoreQuery.filters.push({
      fieldName: "created",
      fieldValue: value,
      operator: ">",
    });
    return this;
  }
  public build(): DatastoreQuery<Task> {
    return this.datastoreQuery;
  }
}

Create DatastoreClient

You can simply create a DatastoreClient with default Datastore configuration, which assumes you are running under Google Cloud environment, e.g., on a Compute Engine. Or pass in your own configured Datastore instance. See @google-cloud/datastore for their documents.

import { DatastoreClient } from '@selfage/datastore_client';
import { Datastore } from '@google-cloud/datastore';

let client = DatastoreClient.create();
let client2 = new DatastoreClient(new Datastore());

Save values

To save values, you can use DatastoreClient's save(), which takes a generated DatastoreModelDescriptor, e.g. TASK_MODEL. The name of the model Task and the id field, because of "key": "id", will be used together as the Datastore key, which also means you have to populate id field ahead of time. Note that only fields that are used by queries are indexed, and the rest are excluded.

import { DatastoreClient } from '@selfage/datastore_client';
import { TASK_MODEL } from './task_model'; // Generated by @selfage/generator_cli.
import { Task, Priority } from './task'; // Generated by @selfage/generator_cli.

async function main(): void {
  let client = DatastoreClient.create();
  // Nothing is returned by save().
  await client.save([{
    id: '12345',
    payload: 'some params',
    done: false,
    priority: Priority.HIGH,
    created: 162311234
  }],
  TASK_MODEL,
  // Can also be 'update' or 'upsert'. See Datastore's doc for what they do.
  'insert');
}

Note that the id field is stripped and converted to Datastore key when saving. If you inspect your Datastore dashboard/console, or query directly from Datastore, you should expect the id field to not be set. The id field will be populated if you use get/query method described below.

Allocate keys/ids

Because we have to populate id field (or whatever field you specified for "key": ...) before saving, you can either use a your own random number generator or use DatastoreClient's allocateKeys().

import { DatastoreClient } from '@selfage/datastore_client';
import { TASK_MODEL } from './task_model'; // Generated by @selfage/generator_cli.
import { Task, Priority } from './task'; // Generated by @selfage/generator_cli.

async function main(): void {
  let client = DatastoreClient.create();
  // The `id` field will be populated in the returned `values`.
  let values = await client.allocateKeys([{
    payload: 'some params',
    done: false,
    priority: Priority.HIGH,
    created: 162311234
  }], TASK_MODEL);
}

Note the field for key has to be of string type and thus we will always store Datastore key as [kind, name]. This decision is opinionated that we don't have to struggle with number vs string when coding, reading or debugging.

Datastore actually allocate ids as int64 numbers, but JavaScript's number cannot be larger than 2^53. Therefore the response from Datastore is actually a 10-based string. We here further convert it to a base64 string to save a bit storage.

Get values

Getting values is straightforward with a list of id.

import { DatastoreClient } from '@selfage/datastore_client';
import { TASK_MODEL } from './task_model'; // Generated by @selfage/generator_cli.
import { Task, Priority } from './task'; // Generated by @selfage/generator_cli.

async function main(): void {
  let client = DatastoreClient.create();
  let values = await client.get(['12345', '23456'], TASK_MODEL);
}

Query with QueryBuilder

QueryBuilders are generated from "queries": ... field. Each of them is named as ${query's name}QueryBuilder, and with ${operator name}${captalized field name}() function(s) which takes a value with proper type as its only argument.

import { DatastoreClient } from '@selfage/datastore_client';
import { TASK_MODEL, TaskDoneQueryBuilder } from './task_model'; // Generated by @selfage/generator_cli.

async function main(): void {
  let client = DatastoreClient.create();
  let taskDoneQuery = new TaskDoneSinceQueryBuilder()
    .equalToDone(true)
    .greaterThanCreated(1000100100)
    // .start(cursor) if you have one to use.
    .limit(10)
    .build();
  let {values, cursor} = await client.query(taskDoneQuery);
}

Note that you need to update the generated index.yaml to Datastore to build those indexes first.

Because query order has already been specified in queries field, you only need to set the values to filter by. And you MUST set all filters, otherwise Datastore might complain the lack of a corresponding composite index.

Delete values

Simply providing a list of id.

import { DatastoreClient } from '@selfage/datastore_client';
import { TASK_MODEL } from './task_model'; // Generated by @selfage/generator_cli.
import { Task, Priority } from './task'; // Generated by @selfage/generator_cli.

async function main(): void {
  let client = DatastoreClient.create();
  await client.delete(['12345', '23456'], TASK_MODEL);
}

Transaction

DatastoreClient also acts as a factory to create transactions, which then can do all operations above but in a transaction. Finally you'd need to commit it.

import { DatastoreClient } from '@selfage/datastore_client';

async function main(): void {
  let client = DatastoreClient.create();
  let transaction = await client.startTransaction();
  // await transaction.save([{}], TASK_MODEL, 'insert');
  // let values = await transaction.allocateKeys([{}], TASK_MODEL);
  // let values = await transaction.get(['12345', '23456'], TASK_MODEL);
  // await client.delete(['12345', '23456'], TASK_MODEL);
  // let {values, cursor} = await transaction.query(taskDoneQuery);
  await transaction.commit();
}

Design considerations

We choose to define datastore field inside message because any change of message must be also reflected in the generated DatastoreModelDescriptor and QueryBuilder in one PR/git commit, to make sure fields are properly indexed. Otherwise, they might not be excluded from indexing or composite indexes might need to be back-filled.